SigLIP 400M

Training compute

4.9×10²¹ FLOP

Parameters

400M

Published

Mar 27, 2023

SigLIP 400M is an AI model developed by Google DeepMind (United States), first published in March 2023. It works in the vision domain, on tasks such as image classification and image embedding.

Training it took an estimated 4.9×10²¹ FLOP of compute (estimation method: hardware,operation counting). The model has 400,000,000 parameters. It was trained on roughly 6.7T datapoints. Training ran on 32 Google TPU v4 for about 120 hours.

Access: Open weights (unrestricted). Its weights are openly available. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Google DeepMind
Country of organization: United States
Domain: Vision
Task: Image classification, Image embedding
Training compute: 4.9×10²¹ FLOP
Compute estimation method: Hardware, Operation counting
Parameters: 400,000,000
Dataset size: 6.7T
Training hardware: Google TPU v4
Chips used: 32
Training time: 120 h
Chip-hours: 3.8K
Training power draw: 21.7 kW
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Epoch confidence: Confident